Pharma tools & lab automation · Chemical synthesis AI

b12 Labs

A two person, then three person, chemistry lab in San Francisco building AI agents that plan chemical reactions and hand the resulting protocols to robots. Founded in 2025 out of Y Combinator, it has no clinical or regulatory record yet, only a paper, a batch of unnamed pharma pilots, and a small seed round.

A lab named for chemical intelligence

Andres Camilo Marulanda Bran and Zlatko Jončev met inside a room with an ambitious name: the Laboratory of Artificial Chemical Intelligence, part of EPFL’s chemistry department. Bran was a PhD student there under Philippe Schwaller. Jončev, who had just finished a PhD in organic chemistry at the University of Basel, was doing a postdoc in the same lab. According to their Y Combinator company profile, a YC partner visited the EPFL campus, and the two decided to apply together.

“We decided to apply after we had a YC partners visit our campus at EPFL,” the founders wrote. “Experience was very intense, a lot of work and great advice on how to build a startup.”

b12 Labs, which also appears in Y Combinator’s listings under the shorter name b-12, joined the accelerator’s Summer 2025 batch. By August 2026 a job board tracker counted three employees; a funding database put it at two. This is still a very small company.

The problem: chemistry is the bottleneck

The founders’ own pitch, repeated across their Y Combinator launch post and their website, is blunt about where drug discovery loses time. It takes more than ten years and over two billion dollars to bring a new drug to market, they say, and only two in ten thousand compounds that enter the pipeline make it out the other end. Chemical synthesis, the step where a proposed molecule actually gets made in a lab, is one of the slowest parts of that chain.

Their diagnosis has three parts. Software for chemists is fragmented, with different tools covering different pieces of the pipeline and little integration between them. Lab robots, the multimillion dollar systems from vendors like Chemspeed and Opentrons that can run reactions automatically, are underused because programming them requires specialists that are hard to find. And the actual work of designing good experiments and iterating through trial and error in the lab simply takes time no matter how good the software is.

b12 Labs’ answer is a stack of AI agents rather than a single tool. Selenium takes a target molecule described in plain language and returns ranked synthesis routes, broken into steps with a confidence score attached to each. Palladium takes a chosen reaction and designs a plate of catalyst, ligand, base and solvent combinations to screen in one shot, laying out the wells itself. Cobalt then translates an approved set of conditions into a protocol that a Chemspeed or Opentrons robot, or another automation platform, can actually execute, without the customer writing code.

The founders’ own record

The clearest outside evidence about either founder is Bran’s. He is the first author of “Augmenting large language models with chemistry tools,” the paper describing ChemCrow, an AI chemistry agent published in Nature Machine Intelligence on May 8, 2024, with Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D. White and Philippe Schwaller as co-authors. Per PubMed, the paper reports that ChemCrow autonomously planned and executed the syntheses of an insect repellent and three organocatalysts, and helped discover a novel chromophore, by combining GPT-4 with eighteen expert-built chemistry tools. An EPFL news article on the work quotes Schwaller explaining the agent’s name: “Because crows are known to use tools well.”

b12 Labs’ own materials describe this work as the first AI agent that can autonomously think and make molecules in a robotic lab, citing more than a thousand citations for the paper and a best paper award at the NeurIPS AI for Science conference. The citation count varies across the company’s own listings, from 750 to 1,100 to over 1,000, suggesting a figure that keeps climbing rather than a fixed one, but no independent source gives a single authoritative count. A YouTube video from 2023 does independently label an early preprint of the work “2023 Best AI Paper,” consistent with, though not full confirmation of, the specific award the company claims.

Jončev’s record traces further back. A biography page from the University of Basel’s Sparr research group, where he did his PhD from 2018 to 2022, says he represented Serbia at the 44th International Chemistry Olympiad in Washington, DC, and received an Honorable Mention from the American Chemical Society in 2012, after competing regularly at Serbian national chemistry competitions in high school. The same page says he worked as a Medicinal Chemistry Intern at Roche in Basel during his bachelor’s studies. That page independently supports the company’s claim that Jončev worked at Roche and competed at the IChO level. It does not itself state the specific claim that he was a three time national champion, though it describes him as a regular award winner in those competitions. A conference exhibitor listing separately describes Jončev as ex-Chemspeed, matching the company’s claim that he worked at a leading lab robotics company.

What is proven, and what is still claimed

Read plainly: the technical pedigree behind b12 Labs is real and independently documented, especially Bran’s authorship of ChemCrow and Jončev’s chemistry background. What is not yet independently documented is the company itself in action: no named customer, no filed Form D, no patent, and no third party account of the pilot results the company describes.

What to watch

• Whether a Form D or an amended one appears on EDGAR under a b12 Labs or B-12 corporate name as the reported seed round, or a later one, gets papered.
• Named pharma or biotech customers moving from pilot to paid, ongoing contracts.
• Any peer reviewed or third party account of the “full conversion in a single attempt” result the company describes.
• Team size, which multiple sources put at two to three people as of mid to late 2026.

In their words

Related companies

• PostEra: Also in Drug Discovery and Delivery. An AI chemistry company that gave its most famous work away for free
• Convexia: Also in Drug Discovery and Delivery. A two person, Y Combinator backed startup that wants AI agents to do the job of a pharma business development team: find overlooked drugs, judge them, and eventually
• Numerion Labs: Also in Drug Discovery and Delivery. Formerly Atomwise
• Lavo Life Sciences: Also in Drug Discovery and Delivery. A Y Combinator startup that sells computer simulations predicting how a drug molecule will crystallize, a step pharma companies otherwise learn by trial and error in the

Sources

1. Public recordAugmenting large language models with chemistry toolsNature Machine Intelligence · May 8, 2024
2. Public recordPubMed entry, PMID 38799228PubMed · accessed Sep 25, 2026
3. IndependentAn AI leap into chemical synthesisEPFL news, Nik Papageorgiou · May 21, 2024
4. IndependentJončev, Zlatko, Research Group Sparr biographyUniversity of Basel · accessed Sep 25, 2026
5. IndependentAndres M. Bran OpenReview profileOpenReview · accessed Sep 25, 2026
6. IndependentEPFL People page, Andres Camilo Marulanda BranEPFL · accessed Sep 25, 2026
7. IndependentEPFL People page, Zlatko JončevEPFL · accessed Sep 25, 2026
8. Independentb-12 Labs funding and team profileAngelsRound newsletter, Ashher · Oct 1, 2025
9. Independentb-12, Funding, Team and InvestorsStartup Intros · Jul 13, 2026
10. Public recordEDGAR company name search, “b12”SEC · accessed Sep 25, 2026
11. Companyb12 Labs: Making hard molecules for Pharma and BiotechY Combinator company profile · accessed Sep 25, 2026
12. CompanyLaunch YC: b-12, Chemistry Copilot for Pharma and BiotechY Combinator launch post · accessed Sep 25, 2026
13. Companyb12 labs websiteb12-labs.com · accessed Sep 24, 2026
14. Companyb12 Labs, exhibiting at Future Labs Live USATerrapinn · accessed Sep 23, 2026
15. Companyb12 Labs company facts and job listingFreehire · Aug 18, 2026